Information Leakage Research Articles (Page 1)

Multifunctional Janus-like nano-fibrous film enables time resolved anti-counterfeiting encryption and UV monitoring.

This paper focuses on the problem of preventing information leakage in neural networks, i.e., assuming that attackers have obtained intermediate-layer features of a neural network, and preventing attackers from inverting these features to the input with private information. We propose a generic method to slightly revise each arbitrary traditional neural network into a multiary-valued rotation-equivariant neural network (RENN) for preventing information leakage. Specifically, we convert real-valued features in the network into multi-ary features, and each element in the feature vector is a multi-ary number. We hide the input information into a certain phase of the multi-ary feature, and rotate the multi-ary feature for attribute obfuscation in the encryption process. The rotation axis and angle can be considered as the private key. In this way, even when attackers have obtained network parameters and intermediate-layer features, they still cannot extract input information without knowing the rotation information. More crucially, the encryption operation does not damage the spatial correlations between features, so that the encrypted features can be easily processed by convolution operations in the neural network without difficulties. In order to implement successful encryption and decryption, the RENN is designed to satisfy the rotation equivariance property. To this end, we propose a set of rules to revise classic operations in the neural network to ensure the rotation equivariance property. Besides, we prove that the $d$d-ary RENN is downward compatible with the $d^{\prime }$d'-ary RENN when $d^{\prime }< d$d'<d. In experiments, the RENN significantly boosts the capacity of preventing information leakage, yet with only mild degradation of classification accuracy, compared to traditional neural networks. Besides, the computational cost is much less than the homomorphic encryption.

To defend the inference attacks and mitigate the sensitive information leakages in Federated Learning (FL), client-level Differentially Private FL (DPFL) is the de-facto standard for privacy protection by clipping local updates and adding random noise. However, existing DPFL methods tend to make a sharp loss landscape and have poor weight perturbation robustness, resulting in severe performance degradation. To alleviate these issues, we propose a novel DPFL algorithm named DP-FedSAM, which leverages gradient perturbation to mitigate the negative impact of DP. Specifically, DP-FedSAM integrates Sharpness Aware Minimization (SAM) optimizer to generate local flatness models with improved stability and weight perturbation robustness, which results in the small norm of local updates and robustness to DP noise, thereby improving the performance. To further reduce the magnitude of random noise while achieving better performance, we propose DP-FedSAM-$\operatorname{top}_{k}$topk by adopting the local update sparsification technique. From the theoretical perspective, we present the convergence analysis to investigate how our algorithms mitigate the performance degradation induced by DP. Meanwhile, we give rigorous privacy guarantees with Rényi DP, the sensitivity analysis of local updates, and generalization analysis. At last, we empirically confirm that our algorithms achieve state-of-the-art (SOTA) performance compared with existing SOTA baselines in DPFL.

This study investigates the application of cloud computing technology in the financial cost accounting of industrial enterprises, with the goal of optimizing asset structure and enhancing management capabilities. To address challenges such as low accuracy, inefficiency, and data leakage in traditional accounting practices, the study proposes an integrated approach combining cloud computing with a support vector machine (SVM) algorithm. Leveraging SVM, this study develops a multi-user cloud-based financial data privacy protection and classification model. The model first preprocesses financial data and secures it using a distributed two-trapdoor public-key cryptosystem, enabling safe data transmission to the cloud platform. Multi-user key management is handled by a trusted third-party key distribution center to minimize communication overhead and ensure privacy. The SVM model is extended to operate within the encrypted domain, utilizing user public keys to prevent the leakage of sensitive information.

The use of information devices on the Internet poses security risks, such as the leakage of personal information and account hijacking. One of the primary causes of these issues is the use of weak passwords and password leakage. However, if users possess knowledge about password strength, they can prevent the creation of weak passwords. In Japan's primary education system, students use information devices to distribute class materials and share contacts. Therefore, providing information security education to young people is of utmost importance today. On the other hand, there is a field known as gaming education. Gaming education is an instructional method that incorporates games into learning to enhance students' understanding and motivation. In this study, we developed an educational information security game, Mallory in Secured Office, which runs on smartphones and is designed for young users. The game is structured as an escape game set in a corporate office. The objective is to escape from the office while encountering information security challenges that could arise in such an environment. We evaluated the game with 22 participants using completion time measurements and a questionnaire survey. The results showed that players were able to engage with the topic of password management, and survey responses indicated increased interest and understanding of proper password handling

Stock price modeling under privacy constraints presents a unique challenge at the intersection of computational economics and machine learning. Financial institutions and brokerage firms hold valuable yet sensitive data that cannot be centrally aggregated due to privacy laws and competitive concerns. To address this issue, we propose a novel Fast-Converging Federated Learning (FCFL) framework that enables decentralized and privacy-preserving stock price modeling. FCFL employs a dual-stage adaptive optimization strategy that dynamically tunes local learning rates and aggregation weights based on inter-client gradient divergence, accelerating convergence in heterogeneous financial environments. The framework integrates secure aggregation and differential privacy mechanisms to prevent information leakage during communication while maintaining model fidelity. Experimental results on multi-institutional stock datasets demonstrate that FCFL achieves up to 30% faster convergence and 2.5% lower prediction error compared to conventional federated averaging approaches, while guaranteeing strong ε-differential privacy. Theoretical analysis further proves that the framework attains sublinear convergence in O(logT) communication rounds under non-IID data distributions. This study provides a new direction for collaborative financial modeling, balancing efficiency, accuracy, and privacy in real-world economic systems.

ABSTRACT A large literature in accounting and finance has examined the effect of earnings announcements on a firm's own stock price, the effect of delays in announcements, and the effect of a firm's earnings announcements on peer firms as mutually exclusive events. However, it is an open question how these latter two effects interact, that is, whether delayed earnings announcements influence the stock price of peer firms. Hence, this article examines the effect of delayed earnings announcements on the magnitude of information transferred to peer firms. We find that a delay in earnings announcement, conditional on it remaining one of the first announcements in the industry, attenuates the information transfer to peer firms by 24%. The attenuation is robust to several alternative model specifications, clustering, fixed effects, and alternative measures of expected earnings announcement dates. Our tests are consistent with overreaction by peer firms and hence inconsistent with investor inattention. Furthermore, our evidence suggests that the attenuation is not entirely attributable to poor‐quality earnings that are typically associated with delayed announcements. However, we do find that information leakage during the delay period and less industry‐relevant information in delayed announcements are potential explanations for the attenuation in information transfer.

Abstract As a newly emerging pillar industry in the medical field, telemedicine relies on the Internet and other transmission networks to complete the transmission of patient information and obtain the consultation results of telemedicine experts, which greatly improves the guarantee of patients’ lives. At the same time, the secure transmission of medical data is also one of the important standards of telemedicine, because any attack or theft caused by the loss of small details, changes, or information leakage will lead to the direction of treatment, resulting in serious consequences. Therefore, a new digital watermarking scheme for medical images is proposed in this paper, which combines image encryption and watermarking protection techniques. In the watermarking aspect, the Canny operator is used to obtain the self-embedding watermark image which is highly related to the plaintext, and the watermark embedding is realized by NSCT(Nonsubsampled Contourlet Transform), DWT(Discrete Wavelet Transform) transform, Fourier transform, and other operations. An efficient S-Box is constructed by using the generated chaotic sequence and the improved Z-transform, which fully disrupts the arrangement position of image pixels, and uses two-way diffusion to complete the distribution of plaintext information in ciphertext to realize image encryption. Simulation results and related tests show that the algorithm can successfully encrypt color and grayscale medical images, and has good robustness and ability to resist various external attacks.

In the current data-driven digital economy environment, customer lifetime value (CLV) prediction is becoming the basis for enterprises to formulate precise marketing strategies and optimize budget allocation. However, the traditional centralized modeling approach faces challenges of information leakage and regulatory restrictions, while the unpredictability of the market requires the prediction model to have real-time adaptability. Therefore, static modeling cannot meet the requirements of real-time decision-making. Thus, this study proposes an idea for a rolling CLV prediction framework, which combines differential privacy with federated learning, enabling secure prediction of customer value and optimizing dynamic budget allocation in the scenario of multi-store cooperation. By using public data from offline experience stores of Suning, Gome and JD from 2019 to 2023, this study establishes a federated learning framework under the protection of the differential privacy mechanism, captures the dynamic changes in customer behavior using a rolling time window, and designs an intelligent budget allocation strategy based on the CLV prediction results. Experimental results confirm that the proposed framework achieves a 91.2% prediction effect under strict confidentiality restrictions ( = 1.0), with a 53.4% improvement in prediction accuracy compared to traditional strategies, and an increase in the return on investment from 1.9 to 4.2, and a 9.7% increase in customer retention rate. This study provides an innovative privacy-protecting solution in a multi-party cooperation scenario, with significant theoretical value and practical significance, and helps to regulate the interpretability development of retail enterprises' marketing.

The nnU-Net framework has played a crucial role in medical image segmentation and has become the gold standard in multitudes of applications targeting different diseases, organs, and modalities. However, so far it has been used primarily in a centralized approach where the collected data is stored in the same location where nnU-Net is trained. This centralized approach has various limitations, such as potential leakage of sensitive patient information and violation of patient privacy. Federated learning has emerged as a key approach for training segmentation models in a decentralized manner, enabling collaborative development while prioritising patient privacy. In this paper, we propose FednnU-Net, a plug-and-play, federated learning extension of the nnU-Net framework. To this end, we contribute two federated methodologies to unlock decentralized training of nnU-Net, namely, Federated Fingerprint Extraction (FFE) and Asymmetric Federated Averaging (AsymFedAvg). We conduct a comprehensive set of experiments demonstrating high and consistent performance of our methods for breast, cardiac and fetal segmentation based on a multi-modal collection of 6 datasets representing samples from 18 different institutions. To democratize research as well as real-world deployments of decentralized training in clinical centres, we publicly share our framework at https://github.com/faildeny/FednnUNet.

Assessing information leakage risks in Japanese enterprises: strategic management and corporate attributes

Background: Coronary artery disease (CAD) remains a leading cause of morbidity and mortality. Early diagnosis reduces adverse outcomes and alleviates the burden on healthcare, yet conventional approaches are often invasive, costly, and not always available. In this context, machine learning offers promising solutions. Objective: The objective of this study is to evaluate the feasibility of reliably predicting both the presence and the severity of CAD. The analysis is based on a harmonized, multi-center UCI dataset that includes cohorts from Cleveland, Hungary, Switzerland, and Long Beach. The work aims to assess the accuracy and practical utility of models built exclusively on routine tabular clinical and demographic data, without relying on imaging. These models are designed to improve risk stratification and guide patient routing. Methods and Results: The study is based on a uniform and standardized data processing pipeline. This pipeline includes handling missing values, feature encoding, scaling, an 80/20 train–test split and applying the SMOTE method exclusively to the training set to prevent information leakage. Within this pipeline, a standardized comparison of a wide range of models (including gradient boosting, tree-based ensembles, support vector methods, etc.) was conducted with hyperparameter tuning via GridSearchCV. The best results were demonstrated by the CatBoost model: accuracy—0.8278, recall—0.8407, and F1-score—0.8436. Conclusions: A key distinction of this work is the comprehensive evaluation of the models’ practical suitability. Beyond standard metrics, the analysis of calibration curves confirmed the reliability of the probabilistic predictions. Patient-level interpretability using SHAP showed that the model relies on clinically significant predictors, including ST-segment depression. Calibrated and explainable models based on readily available data are positioned as a practical tool for scalable risk stratification and decision support, especially in resource-constrained settings.

ABSTRACT Privacy amplification (PA) is a crucial step in Quantum Key Distribution (QKD). The original key with the risk of information leakage is transformed into the final key with information theory security through PA. The speed of PA limits the overall rate of QKD. This article proposes a novel approach of PA. This method connects Square Hash (SQH) and number‐theory transform Hash (NH) in series and utilizes number‐theoretic fast Fourier transform (NTT) to reduce its computational complexity to . This paper provides a security proof based on information theory. To implement the scheme based on field programmable gate arrays (FPGA), this paper designed a new architecture to support it. Several optimization techniques of the PA scheme have been implemented to accelerate computation speed and reduce resource consumption, including parallel NTT computation unit and pipelined carry unit, etc. This architecture is also applicable to other similar PA scheme. The experiment result achieves a higher throughput than the compared FPGA‐based works. After further optimization of the architecture, the clock frequency reaches 250 MHz, and the throughput exceeds 10 Gbps, achieving more than 6 times higher than the fastest PA solution among the compared works.

The article examines the wave and corpuscular nature of quantum objects using the example of the Mach–Zehnder interferometer and discusses the possibility of the so-called “supertunnel effect”. It is shown how the behavior of a photon in an interferometer is determined not by switching between a wave and a particle, but by the preservation or loss of coherence of its quantum amplitude. Key mechanisms are analyzed: formation of superposition at the beam splitter, interference from coherent amplitude recombination, decoherence induced by path information leakage, and recovery of interference in quantum-eraser schemes. Analogous phenomena for electrons, neutrons, atoms and large molecules are discussed, with attention to dominant decoherence sources (collisions, thermal radiation, internal degrees of freedom) and the shrinking de Broglie wavelength of massive objects. The influence of mass, momentum and barrier parameters on tunneling probability is treated, and practical strategies to enhance tunneling (barrier engineering, resonant tunneling, collective effects, and reducing effective mass) are outlined. The conclusion is that quantum laws are universal: wave-like and particle-like manifestations depend on the experimental context and coherence preservation rather than an intrinsic conversion of the object. The concept of “supertunneling” is framed as potentially realizable only if decoherence and exponential suppression can be overcome, with suggested routes for experimental pursuit.

Dual-emission center ratiometric fluorescence probes based on biomass carbon dots for metal ions detection in water and three-level anti-counterfeiting.

The effect of information leakage in a dynamic Cournot duopoly model with bounded rationality

Information leakage detection through approximate Bayes-optimal prediction

ABSTRACT In order to address the data privacy issues in real‐time collaboration in P2P virtual scenes, this paper proposes a privacy protection transmission framework based on the Brakerski Fan Vercauteren homomorphic encryption scheme. The proposed framework allows for direct computation of ciphertext, ensuring end‐to‐end privacy protection while maintaining acceptable real‐time performance. We have built a high‐performance experimental platform consisting of an Intel Xeon Gold edge server with 128 GB of memory, an NVIDIA RTX A6000 GPU, and a Docker based Ubuntu environment that integrates TensorFlow and PyTorch for encryption performance testing and behavioral modeling. The experimental results show that homomorphic encryption supports addition and multiplication operations in ciphertext form, effectively preventing sensitive information leakage. Compared with AES, the BFV based scheme shows higher average latency, but reduces privacy leakage by over 93%. In four real‐world collaboration scenarios, the leakage rate decreased from 86.25% to 5.5%. When the number of nodes increased from 10 to 200, the system throughput decreased from 43.2 to 28.4 MB/s, while the CPU utilization increased from 68% to 87%, indicating a reasonable trade‐off between computation and security. In addition, the integrated edge offloading mechanism reduces encryption latency from 180 to 95 ms, an increase of 36%, while maintaining complete ciphertext confidentiality during processing. These findings validate that the proposed BFV based P2P collaboration framework achieves advanced privacy protection and real‐time scalability, providing a feasible and secure solution for remote design, distributed virtual training, and other privacy sensitive applications. Future research will focus on optimizing lightweight encryption and adaptive scheduling mechanisms for wider deployment in heterogeneous P2P environments.

Multimodal luminescence materials exhibiting multiple stimulus responses are highly favored in optical anti-counterfeiting and information encryption applications. However, this static luminescence under fixed stimuli remains vulnerable to replication. The development of dynamic multicolor luminescent materials offers an effective solution, yet integrating multidimensional dynamic luminescence within a single material remains challenge. Here, this work introduces Tb3+ capture centers into the self-activated luminescent host MgGa2O4-featuring an alternating layered structure and abundant defects-to construct efficient energy transfer channels. This design enables not only static multicolor luminescence dependent on concentration and interplanar spacing, but also, for the first time, stable dynamic multicolor luminescence modulated by four independent dimensions: excitation wavelength, time, temperature, and pressure. In particular, the time domain reveals dynamic photoluminescence with tunable evolution rates, as well as visible-near-infrared dual-band persistent luminescence. These unique optical properties provide strong potential for advanced anti-counterfeiting and visual temperature/stress sensing. Moreover, this work proposes a 4D coupled dynamic encryption system that integrates self-destruction protection and memory fault-tolerance, thereby greatly reducing the risk of information leakage. Combined experimental and theoretical analyses further elucidate the underlying mechanisms, opening new avenues for the design of multidimensional dynamic multicolor luminescent materials.

This paper introduces a novel method that leverages multiple random fusion (RMF) based image encryption techniques to address the demands of high-security applications. In cloud-based privacy-preserving systems for gray-level medical im- ages for health care combine the power of encryption and cloud storage capacity. Inspired by ancient practices of safeguarding wealth by burying it deeply and con- cealing the location as secret keys, this scheme is designed to be scalable. The number of keys required is decided by specific security needs, considering computa- tional resource and encryption time constraints. The method begins by randomly shuffling image blocks or pixels using secure keys or nonlinear transformations such as chaotic maps. Following this, random fusion operations are performed, including shadowing using high-energy signals and adjusting variance to achieve a uniform probability density function. This approach ensures that information leakage is prevented. It is tested for a highest number of key values of 280000 or approximately 1024000, with comparable values of statistical performance parameters. This high key space enhances robustness against brute-force attacks and can be integrated with other encryption methods to provide a general, scalable, and highly secure solution for applications where security is paramount.